Min Zhang posted on Wednesday, October 05, 2016 - 7:57 am
Dear Dr Muthen,
We would like to generate factors scores with probability weights. The means of factors were constrained to zero. We then merged the factor scores back to Stata files. Using Stata, We found that without using weights, the means of factors were close to zero. But if we applied weights, the mean of one of the factors was not zero any more. We would like to use factors scores which have the mean as zero when weights are applied.
This problem was shown on F2.
Mplus syntax: Variable: Names are decidel decided decidev beatout beatnegl beatargue beatnosex serial poolwgt; Missing are all (-9999) ; IDVARIABLE ARE serial ; WEIGHT ARE poolwgt ; CATEGORICAL ARE decidel decided decidev beatout beatnegl beatargue beatnosex ; Analysis: TYPE IS general ; ESTIMATOR IS WLSMV ;
Model: F1 by decidel decided decidev ; F2 by beatout beatnegl beatargue beatnosex ; [F1@0] ; [F2@0] ;
Mplus samstat output: Means F1 F2 ________ ________ 1 0.034 -0.098
Stata output: Mean of factors without weights: F1: .0337763 F2: -.0976396
with probability weights: F1: -.0191061 F2: -.1505213
The Mplus sampstat output with means and variances for the estimated factor scores does not take sampling weights into account. Sampling weights are only used in the model estimation and influence the estimated factor scores only indirectly via the model parameter estimates. You can create the weighted means and variances for the estimated factor scores outside Mplus.